New type of sensor holds promise for machine vision

A technology from InVisage offers an adjustable bandgap as well as a greater sensitivity to light and a wider dynamic range than CMOS image sensors.

Ann R. Thryft, Contributing Technical Editor -- Test & Measurement World, 6/1/2011 12:00:00 AM

This is an expanded version of the article that appeared in the June 2011 Machine-Vision & Inspection Test Report

A new image-sensor technology targeted at consumer camera phones has characteristics that may prove beneficial for machine-vision cameras. Invented by InVisage, the QuantumFilm technology offers an adjustable bandgap as well as greater sensitivity to light and a wider dynamic range than the CMOS image sensors that are being increasingly used in machine vision. An associated technology, QuantumShutter, eliminates rolling shutter.

Although costlier and slower than most CMOS image sensors, CCDs are still used heavily in electronics inspection because of their higher light sensitivity, wider dynamic range, and lower shutter leakage that results from their global shutters (Ref. 1). The rolling shutters of most CMOS sensors produced for consumer markets cause high shutter leakage, creating streaks on video of moving objects on a production line, while lower pixel fill factors causes lower light sensitivity. Neither CCD nor CMOs sensors typically include an adjustable bandgap, a feature that allows a sensor to be “tuned” or optimized to specific wavelengths, including visible and non-visible light. Many consider an adjustable bandgap to be a desirable feature for machine-vision applications, especially for PV (photovoltaic) solar inspection, since it allows a sensor to be optimized for UV (ultraviolet), NIR (near-infrared), or SWIR (short-wave infrared) wavelengths (Ref. 2).

The InVisage technology is a new class of semiconductor material based on nanometer-sized quantum dots applied as the top layer on a standard CMOS wafer. The process was designed to be easily integrated with silicon sensor manufacturing, said Michael Hepp, InVisage’s director of marketing.

“The first major problem with silicon sensors is the fact that the top layer of each pixel is usually metal, which light rays have difficulty penetrating. This can bring down pixel fill factor by up to 50% with decreasing pixel size,” Hepp said. “The second major problem with silicon is its quantum efficiency, or electrical sensitivity to light, which is generally 40 to 60%.” Because a regular, FSI (front-side illumination) CMOS sensor pixel’s photodiodes are buried in silicon under layers of metal, about 50% of light is lost before it can be detected by the photodiodes.

“The quantum efficiency of our material is 80 to 90%, almost double that of silicon,” said Hepp. “Moving the photodiodes to the top layer of film nearly doubles that efficiency again by exposing all of the chip’s top layer, covered with our film, to light, thus allowing 100% pixel coverage. As a result, the light absorption of our film can be nearly four times that of silicon.”

The use of BSI (backside illumination) in standard CMOS image sensors has dramatically improved the fill factor problem, but BSI doesn’t improve quantum efficiency, and it costs more because it requires more steps. QuantumFilm can make use of a foundry’s lower-cost wafers made with older processes, since it can be painted on top of the sensor and requires no special or advanced manufacturing techniques.

For example, to make 1.1-micron pixel sensors, the QuantumFilm method can use a 110-nm standard process on 8-in. wafers, while a standard CMOS sensor must use a 65-nm copper process on 12-in. wafers. Hepp said that the QuantumFilm sensors designed for camera phones will have a 1.4-micron pixel size, and next-generation devices are planned at 1.1 micron.

In a traditional pixel, the “well” that stores data collected by the photodiode is inside the silicon, said Hepp. The bigger the well, the wider the dynamic range of every pixel.

“As you make the pixel smaller, that well size also gets smaller, since it’s a capacitor,” he said. “We can have a much larger well, since we’re not trading off space inside the pixel. That’s because there’s more room inside the silicon, since we’re not putting the photodiodes there. In a 1.4-micron pixel, we can achieve a 12,000-electron well, compared to a 4000- to 5000-electron well in a standard CMOS image sensor at that pixel size, thus increasing dynamic range.”
Because quantum dots are extremely small, their size can be changed to adjust the film’s bandgap, thus tuning or optimizing the sensor to specific wavelengths, said Hepp. “For visible light, the quantum dots are under 5 nm in diameter when packed together,” he said. “A larger diameter, such as more than 5 nm, lets you optimize the sensor for NIR and SWIR wavelengths. SWIR and NIR sensors, however, are usually more expensive because they’re currently made from gallium arsenide.”

A key sensor feature needed for inspection applications is a global shutter, which requires a storage node, said Hepp. A traditional CMOS sensor doesn’t have memory, but reads out data right away after receiving it. A global shutter puts a storage node inside the pixel so it can be read out later. “But that storage node must be shielded from light, which isn’t possible with a standard BSI CMOS sensor pixel, so you sacrifice fill factor,” he said. “There’s also not enough space to contain both the storage node and the photodiode inside the same pixel.” These problems don’t exist with the QuantumShutter option.

“InVisage’s quantum film technology can improve the CMOS sensor,” said Tom Hausken, director of photonics and compound semiconductors at Strategies Unlimited. “The big question you have to ask with a new technology is whether it is destined only for high-volume applications. Sensors for machine vision don’t have to be the smallest or the cheapest, so a new technology has to improve the performance in some way.”

TSMC is expected to sample chips made with the process this summer, said Hepp. T&MW

REFERENCES